From: Knutaf <knutaf@...>
Date: 2009-07-23T23:14:18+09:00
Subject: [ruby-core:24529] Re: Proposal: savepoints

I have code doing exactly #2. The same magic I used on Linux (not fork, but
sending all the fds to the new process) doesn't work as is on Windows. I had
to open a bug for it (http://redmine.ruby-lang.org/issues/show/1644), but if
it's fixed, it ought to work in the same way.

-----Original Message-----
From: Urabe Shyouhei [mailto:shyouhei@ruby-lang.org] 
Sent: Thursday, July 23, 2009 7:02 AM
To: ruby-core@ruby-lang.org
Subject: [ruby-core:24526] Re: Proposal: savepoints

Roger Pack wrote:
>> sp = ObjectSpace.savepoint
> 
> You could implement this using fork...I think?

Only when you have a fork system call, yes.

I guess what Yehuda really need is a In-Service Software Update; where a
process image should be refreshed while its in-service sockets to remain
open.
 What is difficult here is that a socket entity is in kernel, not in that
process in question.  So you have to whether (1) refresh that process image
in a magical way, or (2) spawn a new process and send sockets to it in a
magical way.  Anyway some magic cannot be avoided, and a fork system call of
POSIX semantics is a kind of magic used in (2).  Of course, such a magic is
not always available on Windows.